QACNN: Query-based Attention CNN

Implementation of my research:
Query-based Attention CNN for Text Similarity Map (https://arxiv.org/abs/1709.05036)

6 folders and 1 main script should be included :

• preprocess

  • plot2vec.py
  • qa2vec.py

• word_vec

  • register.json
  • glove.42B.300d.json

• raw_data

  • plot
    • IMDB_KEY.split.wiki
  • question
    • qa.json

• output_data

  • plot
  • question

• utility

  • utility.py

• model

  • MODEL.py

• main.py

usage

1. put the data given by MovieQA into raw_data folder (including IMDB_KEY.split.wiki and qa.json)
2. download GloVe word2vec and put it into word_vec folder (GloVe: https://drive.google.com/file/d/0B3UsyrHYzsTzekMtbndMZXB1Ync/view?usp=sharing)
3. run 2 scripts (plot2vec.py & qa2vec.py), and all the preprocessed data will automatically be saved to output_data folder
4. run main.py